hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Purtell <apurt...@apache.org>
Subject Re: Binary API compatibility is not a requirement for any 0.98 release candidate
Date Tue, 04 Feb 2014 03:01:08 GMT
I don't think we should shy away from breaking (binary API) changes until
1.0. Then the expectations of a stable API kick in, so fix what need fixing
before. This was the rationale here. See the discussion around the time
the KV changes went in on dev@ and the issues themselves. Minor changes
like you point out are fine to commit to 0.98 branch ahead of the next RC
but aren't blocking issues for release. This has been well discussed. It
should not be surprising. We did not just "start the discussion". Changing
the rules in the middle is not fair to the RM.


On Monday, February 3, 2014, Enis Söztutar <enis.soz@gmail.com> wrote:

> Thanks for starting the discussion.
>
> I think it should be ok to go w/o binary compat. The only downside is the
> extended effort the downstreamers have to bear, since they might have to
> release versions compiled against 96 or 98 separately.
>
> I went over the changes from Alex in client and hbase packages.
>
> It seems that the Scan change is the only one in the client package for
> binary compat for the Public interfaces. We have a patch at HBASE-10460.
>
> HBASE-10324 caused the changes in HTD to break the deprecated methods. But
> it is easy to fix.
>
> This leaves only the KV changes:
>
> http://people.apache.org/~jmhsieh/hbase_jdiff_report-p-0.96-c-0.98/changes.html
>
> KV is changed to private, so there is no point in trying to undo those
> changes I think.
>
> I would +1 documenting that 0.96 and 0.98 are not binary compatible, but in
> the mean time fix the 2 issues in Scan and HTD so that if the users are not
> using KV, they will be binary compat. What do you think?
>
> BTW, I've opened https://issues.apache.org/jira/browse/HBASE-10462 as a
> blocker against 1.0.
>
> Enis
>
>
> On Mon, Feb 3, 2014 at 3:12 PM, Nick Dimiduk <ndimiduk@gmail.com<javascript:;>>
> wrote:
>
> > FYI, our system tests recently discovered the class rename in HBASE-10431
> > also imposes a binary incompatibility for launching mapreduce jobs when
> > specifying 0.98's hbase-protocol.jar on the HADOOP_CLASSPATH.
> >
> >
> > On Mon, Feb 3, 2014 at 4:08 PM, Andrew Purtell <apurtell@apache.org<javascript:;>
> >
> > wrote:
> >
> > > If you would like to change this consensus now, we can do so, and add
> it
> > as
> > > a release criterion. That would require undoing the comparator cleanups
> > and
> > > related breaking changes that went in as HBASE-9245 and subtasks. So
> > let's
> > > not. I am -1 on making a change like this late in the day, after we
> have
> > > already had two RCs and I am hoping to get a third out tomorrow.
> > >
> > > --
> > > Best regards,
> > >
> > >    - Andy
> > >
> > > Problems worthy of attack prove their worth by hitting back. - Piet
> Hein
> > > (via Tom White)
> > >
> >
>


-- 
Best regards,

   - Andy

Problems worthy of attack prove their worth by hitting back. - Piet Hein
(via Tom White)

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message